On 22/02/2021 15:15, Derrick Stolee wrote:
This change looks like a sane change to me. It definitely fixes a leak.
The leak "wasn't hurting anybody" because write_commit_graph() is only
called at most once per process, and the process closes itself out
shortly after. Still, it's good to have good memory hygiene here.
Good to know - thank you! As I become more familiar with git, I'm
beginning to realise that most leaks are unlikely to be much importance
(even though I personally err on the side of fixing any and all issues).
One thing I forgot to mention: in this specific case the leak was
causing a build failure when trying to build git's fuzzers within
oss-fuzz locally*. Specifically the following command would fail (see
also fuzz failure reproduction instructions which describe the setup [1]).
$ python infra/helper.py build_fuzzers --sanitizer address git
As far as I can tell the issue is that: a copy of git built with ASAN is
used to produce the fuzzing corpus as part of the git-specific build
script [2] - the leak warning causes the script to fail. (It's possible
to argue that the build script should disable ASAN's leak checking when
running git, via detect_leaks=0 to reduce the risk of such breakage - I
may try to suggest such a change to oss-fuzz.)
ATB,
Andrzej
* Given that oss-fuzz is building via docker, I would intuitively
suspect that the same issue occurs in automation - I'm not sure how to
verify this myself.
[1]
https://google.github.io/oss-fuzz/advanced-topics/reproducing/#building-using-docker
[2]
https://github.com/google/oss-fuzz/blob/1b0115eefd70491376cf3cb6f88e49632c78ee18/projects/git/build.sh#L37
